BUG #6763: Severe memory leak with arrays and hstore

The following bug has been logged on the website:

Bug reference: 6763
Logged by: Luben Karavelov
Email address: karavelov(at)mail(dot)bg
PostgreSQL version: 9.1.4
Operating system: Debian Linux
Description:

I was trying to migrate a big ER table (user preferences) into a new table.

Old able:
old_prefs (
user_id integer NOT NULL,
name varchar NOT NULL,
value varchar NOT NULL
);

New table:
preferences (
user_id integer PRIMARY KEY,
prefs hstore;
);
The query I have tried to use is:

INSERT INTO preferences
SELECT user_id,hstore(array_agg(name), array_agg(value))
FROM old_prefs
GROUP BY user_id;

But the postgres backend consumed all the available memory (6G free + 4G
swap) and finally was killed by the kernel.

Its 8G RAM machine and here are the memory options from the config:

shared_buffers = 1800MB
temp_buffers = 16MB
work_mem = 64MB
maintenance_work_mem = 256MB
max_stack_depth = 2MB

Finally I have managed to migrate it in batches of 100-200k user ids and
disconnecting after each query in order to free the backend and leaked
memory.

Best regards
Luben